The old laptop. The exercise bike doing service as a coat rack. Two phones in a drawer. The good coat that never quite fit.

Most homes are sitting on a few thousand dollars of that.

It never gets sold because of the boring middle bit. Working out what each thing is worth, writing the listings, posting them, then dealing with someone offering forty dollars for a two hundred dollar item at eleven at night.

ChatGPT Work does that middle bit now. All of it, including the haggling.

The first prompt is below. Run it before you read the rest, because finding out what your junk is actually worth is a strange and slightly addictive feeling.

The 20-second version: Photograph the stuff you never use. ChatGPT prices each item against real recent sales, writes the listings, opens a browser and posts them to Facebook Marketplace, then negotiates with buyers on a schedule. You keep the final yes on any sale. The pricing step is below. The rest is four prompts.

Time: 10 minutes to a full priced list. About 30 to live listings. You need: ChatGPT on a paid plan (Work), and a Facebook account for the posting steps. Payoff: cash out of a spare room, without the tedious middle bit.

Selling secondhand is not hard. It is tedious in four separate ways, and any one of them is enough to make you close the tab.

The old way: you dig through listings to price it. You write a description that does not read like a ransom note. You post it. Then you answer the same three questions from nine people and hold your nerve when someone offers you forty dollars at eleven at night. Most of it never gets sold because step one alone kills the afternoon.

The replacement: ChatGPT Work does all four. It prices against real recent sales, writes the listings, opens a browser to post them, and negotiates on a schedule. You keep the final yes.

Start with the pricing. It earns its keep on its own.

Photograph everything you want to sell. Decent light, plain background, a few angles, and get any labels or model numbers in shot. Then open ChatGPT, choose Work, upload the photos into one chat, and turn web search on with the globe icon.

Paste this.

What you get: every item identified and priced against what things are really selling for, with two numbers each. The "worth" number is your anchor. The "realistic" number is what actually moves it this week. Anything secretly valuable gets flagged so you do not give it away.

If the prices come back suspiciously round, web search is off. Turn on the globe and say "redo this using real current listings." If it stops halfway through a big batch, say "keep going."

Most people get surprised twice. Once by something worth far more than they assumed, and once by something they were about to overprice into never selling.

A price list is not a sale though. What turns it into money is the next three steps: listings written to get tapped, posted for you, and a buyer conversation that runs without you sitting on your phone all evening.

That is what is behind the wall this week, along with four more complete jobs ChatGPT Work does across real apps.

🔒 Unlock the rest of the selling system:

The Selling System, finished. The pricing step tells you what your stuff is worth. These three turn that into money:

The scam filter. The three seller scams that hit everyone, and the prompt that flags a sketchy message before you reply.

Four more complete jobs, each with the app to connect and the starter prompt:

Also in your library: the Computer Use Playbook, the Unattended Job Pack, the App Replacement Pack, the Design Systems Pack, 100 Things You Can Do With Claude and ChatGPT, the Command Codes Pack, and 80+ prompts. Every workflow from every issue, permanently.

Google renamed NotebookLM to Gemini Notebook on July 16 and gave every notebook its own secure cloud computer that can write and run code.

If you have not used it, the idea is simple and better than it sounds. You upload your own documents and it answers only from those, showing exactly where each answer came from. Nothing invented from the wider internet. Upload a pile of contracts, a year of reports, or your own notes, and interrogate them.

The code execution is the new part. Hand it your own spreadsheet and it runs real analysis rather than describing analysis you could do. Rolling out to Ultra and Workspace accounts first, Pro users next (source: Technobaboy).

30 million people use it, mostly for study notes. It is one of the better business tools nobody thinks of as a business tool.

☕ Starbucks is building its own AI software to replace Microsoft and IBM tools. Bloomberg reported the chain is developing in-house replacements for a Microsoft inventory system and an IBM maintenance platform, against a software bill of roughly $400 million a year (source: Bloomberg).

🇨🇳 The largest open AI model ever released landed this month. Moonshot AI shipped Kimi K3 on July 16, around 2.8 trillion parameters with a 1-million-token context window, and it took the top spot on a major coding leaderboard within days (source: Build Fast with AI).

📓 NotebookLM is now Gemini Notebook. Google renamed its research tool on July 16 and gave each notebook a cloud computer for running code, with 30 million users and 600,000 organisations on it (source: SiliconANGLE).

🌏 A 29-country AI governance body launched in Shanghai. The World AI Conference closed July 20 with Xi Jinping's first AI keynote and the founding of the World Artificial Intelligence Cooperation Organization (source: Build Fast with AI).

Quick version: AI agents drove 20% of global holiday orders last year, $262 billion worth. AI-referred shoppers went from converting 38% worse than normal traffic to 42% better in twelve months. You are selling with an agent. The buyer increasingly has one too, and it reads every listing in the category before recommending three. Write for it.

Here is a number that should stop you.

During the 2025 holiday season, AI agents drove 20% of global orders. That is $262 billion in sales handled by software shopping on someone's behalf (source: Elogic, citing Salesforce State of Marketing 2026).

Then the number that actually matters.

In March 2025, traffic arriving from AI tools converted 38% worse than normal traffic. Twelve months later, in March 2026, it converted 42% better. An eighty point swing in a year.

Read that twice. AI-referred shoppers went from the worst traffic on the internet to the best, in four quarters.

The volume moved with it. AI-driven traffic to US retail sites grew 393% year on year in the first quarter of 2026. Around 50 million shopping queries now happen inside ChatGPT every day, roughly 2% of everything people ask it (source: Elogic, citing OpenAI Economic Research).

So this week's workflow is one half of a bigger shift. You are using an agent to sell. A growing share of the people on the other side are using an agent to buy, or at least to decide.

That changes what a good listing is. For twenty years you wrote listings for a human skimming a feed. Increasingly you are also writing for a machine that reads every listing in a category, compares them properly, and recommends three.

Most sellers have not noticed. The ones who adjust first get found first.

🔒 The Pro breakdown: what AI buyers actually read and rank, how to write a listing or product page that agents surface instead of skip, and the honest limits of what anyone knows about this yet.

Workflow Library Updates:

NEW Resources:

Each resource lives permanently in your Pro account. Use them whenever you need them.

Your clutter is somebody else's Tuesday afternoon purchase. The only reason it is still in your spare room is the tedious middle bit, and the tedious middle bit is now somebody else's problem.

If you do one thing before next Thursday, photograph five things and run the pricing prompt. Worst case you find out what your old laptop is worth. Best case you have cash by the weekend and a system you can run again every time the cupboard fills up.

👇 What’s behind the paywall:

Pro members deploy AI in their work an average of 5-8x more often than free readers (based on reply data from past issues). The difference is having the exact setup, not the concept.